Globally Aligned Optimization
Hierarchically decomposed by time phase, globally aligned by a shared objective. Not local optimization, not monolithic optimization -- the correct middle ground for tractable, real-time port control.
One Shared Business Target
Every layer, every solver, every replan serves a single measurable goal: minimize total weighted controllable vessel sojourn time.
where: T^ready_v = time vessel v becomes service-eligible T^dep_v = actual departure-ready time of vessel v w_v = priority weight of vessel v
Minimize the total weighted controllable sojourn time across all vessels. This captures only the time the port controls -- from service-eligibility to departure readiness -- not the full arrival-to-departure span.
Pre-Berth: admission delay Service: berth service time Post-Service: release delay
The controllable sojourn decomposes into three mutually exclusive, collectively exhaustive phases. Each phase maps to exactly one optimization layer. No gap, no overlap.
Naive arrival-to-departure includes uncontrollable factors (vessel early arrival, weather delays). We measure only what the port can actually optimize.
w_v encodes commercial priority, vessel size, and contractual commitments. High-priority vessels get proportionally more optimization pressure.
Three phases, three layers. Every second of controllable sojourn belongs to exactly one layer. No optimization gap, no double-counting.
Why This Architecture
Two common approaches fail at port-scale. The correct answer is in the middle: decomposed for tractability, aligned for correctness.
Not Local Optimization
Optimizing each equipment subsystem independently produces local optima that conflict at boundaries. A crane-optimal plan can starve AGVs; an AGV-optimal route can block yard access. Local optimality does not equal system optimality.
Not Monolithic Optimization
A single giant solver across all resources is too large to scale, too slow to run in real-time, and too brittle when any sub-problem changes. Monolithic formulations collapse under operational complexity.
Hierarchical + Globally Aligned
Decompose by time phase, not equipment. Each layer solves a tractable sub-problem while a global coordinator enforces inter-layer consistency constraints. Decomposition gives speed; alignment gives correctness.
MECE Time Decomposition
Each layer owns exactly one phase of the controllable vessel sojourn. Proposals flow forward, evaluations flow backward, creating a closed coordination loop.
Admission delay minimization. Assigns vessels to berths, determines berth start timing, and sequences admission under marine-access and capacity constraints.
Berth service time minimization. Orchestrates STS crane assignment, AGV/truck dispatch, and yard-side container absorption. The most operationally critical layer.
Release delay minimization. Manages departure clearance, berth release sequencing, and resource conflict resolution for outbound vessels.
Variables: x = Pre-Berth decisions y = Service decisions z = Post-Service decisions Output of x shapes feasible region of y. Output of y shapes feasible region of z.
The optimization is not a pipeline -- it is a coupled system where each layer's decisions constrain the next layer's feasible set.
Pre-Berth sends berth assignment proposals and berth start timing to Service.
Service sends projected work-done times and resource forecasts to Post-Service.
Service returns feasibility flags, congestion penalties, fragility scores, and infeasibility cuts.
Post-Service returns berth release times and departure resource conflicts.
Closed loop -- not a one-way pipeline
Coordination Mechanism
The 7-step production operating logic that turns three independent solvers into one coherent system.
1. World State Update
Ingest current vessel positions, berth occupancy, crane status, AGV fleet state, yard density, and departure queue into a unified state snapshot.
2. Pre-Berth Proposes
Pre-Berth layer generates berth assignment proposals and berth start timing based on vessel queue, priority weights, and marine access windows.
3. Service Evaluates
Service layer evaluates each proposal: returns projected work-done time, resource requirements, congestion penalty, fragility score, and feasibility flag.
4. Feedback Returns
If infeasible or fragile, Service sends infeasibility cuts and congestion penalties back to Pre-Berth. Pre-Berth re-proposes with tighter bounds.
5. Plan Commits
Once feasible, the berth plan commits. Service begins execution. Post-Service receives projected work-done times and prepares departure resources.
6. Post-Service Evaluates
Post-Service returns berth release times and flags departure resource conflicts. Conflicts propagate back to the coordinator for resolution.
7. Freeze and Execute
Committed decisions are frozen. Only the remaining horizon is re-optimized on the next trigger. Plan stability is preserved as a first-class objective.
vessel_id
berth_id
planned_berth_start
priority_weight
work_profile
marine_access_window
vessel_id
feasible_flag
projected_work_done_time
projected_qc/agv/yard_requirements
congestion_penalty
fragility_score
vessel_id
projected_departure_ready_time
berth_release_time
departure_resource_conflict
Rolling-Horizon Control
Optimization is not one-and-done. The system continuously re-solves on triggers while preserving committed decisions and plan stability.
Ingest the triggering event into the unified state model.
Lock in-flight operations that cannot be reversed without operational cost.
Re-solve only the uncommitted future window with updated constraints.
Penalize unnecessary churn. Minimize the delta between old plan and new plan.
Pre-Berth re-sequences admission queue
Service re-dispatches cranes and AGVs
Service adjusts routes, signals Pre-Berth
Service reroutes absorption; may delay berth start
Post-Service re-sequences release queue
Post-Service adjusts departure windows
Rule engine + optimization orchestrator
Enforces inter-layer consistency, routes feedback, manages freeze/commit cycle
MILP, CP-SAT, decomposition-based admission scheduler
Berth assignment and admission sequencing under capacity and marine constraints
MILP, CP-SAT, LNS, hybrid heuristics, dispatch policy layer
Crane-vehicle-yard coupled optimization with real-time feasibility repair
Event-driven repair heuristic, local reoptimization
Departure clearance and berth release with conflict resolution
Optimization Layers
Explore each layer's mathematical formulation, solver architecture, coordination contracts, and operational boundaries.
Pre-Berth Optimization
Admission delay minimization through berth assignment, start timing, and vessel sequencing under marine-access and capacity constraints.
Service Optimization
Berth service time minimization through coupled STS crane assignment, AGV dispatch, and yard-side absorption orchestration.
Post-Service Optimization
Release delay minimization through departure clearance, berth release sequencing, and outbound resource conflict resolution.